Let’s drop the object : An experimental study of the implicit object construction in American English (2024)

"... il y a donc dans chaque localité un type linguistique idéal dont toutes les réalisations de fait ne sont que des approximations."

A. Meillet, Les Dialectes Indo-Européens, 1908.

1Since Fillmore’s (1986) inspiring work on zero anaphora in American English, an implicit direct object construction (IOC) has been recognized as such and received much attention not only in the Construction Grammar (CxG) literature (Goldberg 1995, 2001, 2005, 2013; Hoffmann & Trousdale 2013; Kay & Fillmore 1993; Lambrecht & Lemoine 1996; Mittwoch 2005; Rakhilina 2010), but also in formal/generative approaches to syntactic structure (Cote 1996; Cummins & Roberge 2004, 2005; Cyrino & Lopes 2016; Giannakidou & Merchant 1997; Huang 1984; Koulidobrova 2017; Lavidas 2013; Paducheva 1977; Ruda 2014, 2017). In addition, phonologically null objects, their syntactic distribution and semantic/pragmatic characteristics, have been studied by language acquisition researchers as a separate phenomenon, not necessarily as a component oflarger constructions with well-established discourse properties (e.g. Boping 1997; García Mayo & Slabakova 2015; Grüter 2007; Medina 2007; Mykhaylyk & Sopata 2016; Park 2004; Pérez-Leroux, Pirvulescu, Roberge & Castilla 2013).

3English, in its multiple dialects, is traditionally described as a language with strict lexical and/or discourse-based restrictions on the omission of core arguments (subject, direct object, indirect object), which are argued to differ significantly from languages with more flexible argument omission patterns (e.g. Chinese, Ukrainian, Portuguese, inter alia). On the CxG approach, if a transitive verb’s object has no overt lexical content, it is assumed to instantiate a construction that assigns formal features only to a subset of core/non-core arguments, with the exact number of arguments determined by a verb’s thematic grid and/or construction.

5In anticipation of a more detailed discussion in Section 2, there are several important observations that should be made about the examples in (1). First, the interpretation of the implicit objects depends on the verb and context. In addition, we note the following parameters of variation:

6(i) specificity/(in)definiteness: some objects are indefinite/non-specific, some are definite/specific;

7(ii) referentiality: some objects are non-referential (generic), some are referential;

8(iii) deictic reference: some objects are deictic in reference, some are not;

9(iv) count/mass distinction: some objects are interpreted as plural and/or singular, some as mass nouns;

10(v) generality of noun semantics and noun status in a hierarchical semantic network: hypernym vs. hyponym (e.g. games/Space Invaders);

11(vi) antecedent-dependence: some are antecedent-dependent (i.e. lexically constrained by the presence of an antecedent nominal), some are antecedent-free (because no antecedent, pragmatic/deictic or lexical, is available).

12Second, a multitude of parameters associated with implicit objects has provided fodder for interesting discussion in the CxG, which integrates semantic/pragmatic properties into the internal structure of constructions. The problem of (in)exact meaning, along with the problem of semantic recoverability (i.e., the process by which speakers ‘restore’ appropriate and contextually felicitous object semantics) has been particularly prominent. This article is concerned with how interpretations of implicit objects in American English (AE) are shaped by different discourse conditions. Specifically, we focus on whether there are differences in speaker acceptability of the IOC in contexts that provide for a lexical antecedent vs. those that do not.

13We report the results from two experiments that explore context-related acceptability of implicit objects in informal dialogues. Experiment #1 investigates speakers’ acceptability judgments, asking them to use a 7-point scale to rank sentences with implicit objects vis-à-vis their replicas with a lexicalized object. Speakers’ judgments are tested under two types of discourse conditions: with an overt lexical antecedent (definite vs. indefinite) and without an antecedent. The first context favors a referential interpretation for an implicit object, whereas the second favors a nonreferential (generic) interpretation. Experiment #2 uses the same dialogue materials to test speakers’ judgements in a forced multiple-choice task, in which they are presented with implicit/overt object variants of the same construction and are asked to choose only one that they consider to be "the best fit."

14The article is organized as follows. Section 2 provides discussion of the IOC and implicit object semantics in the CxG framework, focusing on the issues of contextual felicity and semantic recoverability. Section 3 introduces the experiments and reports the findings. Section 4 provides general discussion. Section 5 concludes the article.

15Here we provide an overview of the issues relevant to understanding an implicit object construction (IOC). In subsection 2.1, we unpack the contents of an IOC by focusing on the details of its internal structure. Based on Goldberg (2005), two variants of an IOC are introduced – The Implicit Theme Construction and the De-profiled Object Construction. Our goal is to explain how an IOC is analyzed in the CxG and to highlight its semantic/pragmatic characteristics, which account for its conventionalized functions in American English. In subsection 2.2, we focus on the semantic/pragmatic properties of implicit objects and explain how an IOC can have both a non-referential and referential interpretation (sometimes in the same context). The ensuing discussion is relevant for understanding the hypotheses of the study introduced in Section 3.

2.1 An implicit object construction and its variants

16In the CxG, the notion of a construction extends to morphemes, words, phrases and syntactic patterns of various generality. Furthermore, constructions, hierarchically organized and conventionalized in their function, constitute speakers' linguistic knowledge in its entirety. A construction is broadly conceptualized as a (monostratal) symbolic unit: a “conventional pairing of form and function” (Goldberg 2005: 224) and "a set of conditions licensing a class of actual constructs of a language" (Kay & Fillmore 1999:3). As a schematic pattern, an IOC integrates (at least) three levels of linguistic representation: semantic (specifying participant roles generalized over verb classes), syntactic (specifying core and non-core arguments and their formal status within a construction: obligatory/instantiated, obligatory/uninstantiated, optional), and pragmatic (assigning topic/focus functions to core/non-core arguments). Goldberg (2001, 2005) provides an extensive discussion of an IOC, arguing that implicit objects are part and parcel of two types of constructions: (i) the Implicit Theme Construction and the De-profiled Object Construction. Each construction type is motivated by a separate set of empirical facts (see below) and is associated with its own schema. We discuss each construction type in detail in the following subsections.

2.1.1 The Implicit Theme Construction

17The Implicit Theme Construction (ITC) specifies information about semantic verb classes, participant roles generalized over individual verbs from the relevant classes, as well as syntactic relations linked to each participant role.Examples in (2 & 3) illustrate the ITC (all are from Goldberg 2005).

18The set in (2) consists of the ‘caused motion’ verbs, which are further subdivided into three classes based on their lexical semantics: (i) emission verbs: spit, sneeze, blow, leak, etc. (ii) ingestion verbs: eat, drink, etc., and (iii) contribution verbs: contribute, donate, etc. (The implicit objects are lexicalized in parentheses; all of the overt nouns are Goldberg’s suggestions, except in (g-h), which are my own.)

20For (2 a-f), Goldberg observes that the verbs can appear either with or without a directional prepositional phrase, and can appear with or without a theme argument (cf. Pat sneezed (onto the computer screen)/Pat sneezed (her mucus)). The minimal structure for a verb like sneeze (and verbs similar to it) is ‘Pat sneezed.’ This observation is used to make two important arguments. First, the caused-motion verbs provide an empirical challenge to the Argument Realization Principle (ARP) (Grimshaw & Vikner 1993; Rappaport Hovav & Levin 1998, 2000, inter alia). According to the ARP, there must be a requirement that lexical verbs with a complex sub-event structure have “one argument in the syntax to identify each sub-event in the event structure template” (Goldberg 2005: 216). If all the predicates in (2) are assumed to have a complex event structure template, which includes a motion sub-event that entails either a change of location (2a-f) or a change of ownership (2g-h), then the empirical accuracy of the ARP is in doubt and an alternative account must be proposed.

22The Semantics tier in Figure 1 specifies key participant roles in a generalized semantic form: Source, Theme, Direction, with the bold font indicating that only Source must be obligatorily lexicalized and ‘linked’ with a syntactic relation (Subject). The line marked as PRED, along with the parentheses that follow, captures lexical restrictions on the verb classes.

23The Syntax tier specifies the types of argument roles (syntactic relations), which include core and non-core arguments. The vertical lines connecting the participant roles to synactic relations indicate which participant is ‘linked’ with which argument (straight lines indicate obligatory linking and dashed lines optional linking). The symbol 'the null sign’, indicates that no syntactic relation is available for the theme participant and therefore it can remain unlexicalized/implicit (in Goldberg's words, it “may alternatively be contributed solely by the construction” (2005: 228)). What this means is the theme/patient participant must be a component of a proposition expressed by the ITC since it is represented in the Semantics tier. The next subsection will deal with the Pragmatics tier, which is absent in Figure 1.

2.1.2 The De-profiled Object Construction

24The second type of construction proposed by Goldberg (2005) is the De-profiled Object Construction (DOC). This construction is interesting in two respects: (i) it does not seem to be restricted to specific verb classes that instantiate it and (ii) the key characteristics of its Pragmatics tier specify the discourse conditions under which it can be used felicitously.

25The data in (4) illustrate the DOC (from Goldberg (2005: 231), including her notes in square brackets).

26Figure 2 (from Goldberg 2005: 232) provides a schema for this construction type. It is the Pragmatics tier that is of interest and will be relevant for our study.

27The Prag (Pragmatics) tier specifies restrictions on various components of the predicate (Pred) and its participant roles. (In this version of the schema, note that only the theme/patient role is included in Figure 2, and the Semantics tier consists of Pred and generalized semantic roles.) Since the Prag tier assigns topic/focus functions to a predicates' arguments, we can see that this construction assigns no discourse function to a theme/patient role: it is neither Topic (non-topical), nor Focus (non-focal). On the other hand, the Pred, marked as P, is ‘emphasized,’ which means that it is coded as new, possibly focal information by the Prag.

28The état d’affaires that emerges at the Prag results in two important characteristics of the DOC: the verb gains the status of a component loaded with pragmatic information, whereas its theme object argument gets “de-profiled” (or backgrounded). Thus, the no-discourse-function status at the Prag coincides with the no-syntactic-relation status at the Syn (Syntax), marked as Æ in Figure 2. Yet, the theme/patient participant remains lodged in the Semantics tier and therefore is part of a proposition once a specific verb instantiates the DOC. An implicit object is argued to be contributed by the DOC proper and the construction itself is predicted to be felicitous under conditions of "low discourse prominence."

29This insight is formalized by Goldberg (2005:230) as the Principle of Omission under Low Discourse Prominence: “Omission of the patient argument is possible when the patient argument is construed to be de-emphasized in the discourse vis-à-vis the action.” Therefore, we can look at this construction as an adaptation of language structure to specific discourse types, namely those that give salience to a predicate while turning a theme/patient participant into an unlexicalized ‘exile’ by denying it a full ‘residence status’ in the syntactic and pragmatic components. As Goldberg (2005:232) suggests, these discourse types could be shaped by “Grice’s maxim of Quantity (second half) to ‘say no more than is necessary,' there’s motivation to leave these particular arguments out.” In Section 3, we will consider the implications of this analysis for referential vs. non-referential construals of implicit objects.

2.2 Approaches to implicit object semantics

30This section focuses on the semantics/pragmatics of implicit objects. How do language users arrive at the appropriate interpretation for a theme/patient participant in an IOC? Is there a cognitive mechanism that speakers draw on when an IOC is implemented in discourse?

2.2.1 Antecedent-based interpretations for implicit objects

31Let us begin with antecedent-based interpretations. With respect to the data in (1), we pointed out that contexts with implicit objects may or may not include a lexical antecedent, suggesting a difference in how their semantics can be derived. In (5), we present an example with the verb play that clearly shows an antecedent-dependent interpretation for the implicit object (note that RPG = ‘role-playing game’):

32Within the minimal context in (5), it is obvious that the theme participant is ‘RPG videogames.’ Therefore, the availability of an antecedent must trigger a mechanism of anaphoric linking, which plays a significant role in the accuracy of semantic interpretation. However, if we imagine a version of this sentence without the noun ‘RPG videogames,' the implicit object semantics becomes much more open-ended, and could conceivably include nouns such as ‘a musical instrument,’ ‘cards,’ ‘mind games,’ ‘poker,’ ‘violin,’ etc., with some of these possibilities entering hyper-hyponymic relations (e.g. musical instrument/violin, cards/poker). Therefore, contexts without a lexicalized antecedent must rely on a different cognitive process in establishing semantic reference for implicit objects.We look at these context types in the next subsection.

2.2.2 Liverpool won: “highly predictable” implicit objects

33In Section 2.1, we discussed the proposal that some implicit objects are assigned no discourse function because (i) they have no (or minimal) relevance in discourse and (ii) their lexical content is highly predictable, namely, it is so plainly obvious that an attempt to lexicalize such an object creates not only stylistic awkwardness but also degrades pragmatic acceptability of a host sentence.

35Turning to object arguments and the question of why they can remain implicit with some verbs, we first summarize the key ingredients that ensure their semantic recoverability. These ingredients are discussed in various research articles (e.g. Fillmore 1986; Fillmore & Kay 1993; Goldberg 2001, 2005; Rakhilina & Plungian 2010, among others). Here we pull them together under the label ‘semantic recoverability of implicit objects’ and explain their importance with examples in (6) & (7).

36(i) contribution of an IOC proper, which provides for a theme/patient participant in the Semantics tier, ensuring that its interpretation is accessible (even though this semantic role is assigned no formal feature in the Prag and Syn tiers within an IOC);

37(ii) a lexical verb instantiating an IOC and a verb’s thematic grid, which includes (a) verb-specific participant roles (e.g. play: player, ‘object of play,' location/time) and (b) a range of default semantics for some participant roles (e.g. for the theme role designated as ‘object of play,’ default lexical meanings could include ‘musical instrument,’ ‘game,’ etc.);

38(iii) situational semantics, which makes for specific lexicalizations of participant roles in a given sentence: 'The Gators will come to play _ against Florida State.' (cf. example (6)) below;

39(iv) inferential process during which speakers make use of their general world knowledge to interpret a sentence in a given situation (e.g. knowledge of the denotees of the noun phrases 'the Gators' and 'Florida State').

40To illustrate, consider the sentence in (6):

41In order to interpret the implicit object in (6) with precision, we need to know that ‘the Gators’ is a shorthand for a university football team ‘Florida Gators’ in the US, and ‘Florida State’ refers to their rival team ‘Florida State Seminoles.’ Based on this knowledge, the theme role of ‘play’ in (6) can only be understood as ‘a football game’ (not 'a soccer game', 'a hockey game,' etc.). The discourse context for (6) centers around the topic of rivalry between the two university teams and their prior wins and losses in football competitions. Thus, specifying that one team will play a football game against the other would constitute informational overload. Such discourse context seems ripe for using the DOC variant of the implicit object construction –the pragmatic focus shifts entirely to the verb.

42Rakhilina & Plungian (2010) highlight the role of situational semantics in the interpretation of implicit objects, analyzing a sentence that mentions a British soccer team ‘Liverpool’ (here we present an identical example from COCA):

43It is possible to interpret the implicit object of ‘won’ ('the soccer game') with utmost precision if we know who Luis Suárez is (a professional soccer player) and what kind of team ‘Liverpool’ is. (Without this knowledge, only a vaguer guess such as 'some game' would be an option.) Comparing (6) and (7), we observe that the semantics of the noun ‘game,’ presupposed in a theme participant of ‘win’ and ‘play,’ is retained in either interpretation but is necessarily adjusted pragmatically to reflect the meaning of the lexicalized arguments. The inferential process further fine-tunes the meaning for each object: 'football game' vs. 'soccer game.' Once again, we encounter a discourse context in which object semantics is "highly predictable," and the use of a DOC becomes pragmatically felicitous. These examples illustrate how implicit object semantics can be cogently derived solely on the basis of contextual/discourse cues, with an interpretation as precise as in the examples with an explicit antecedent.

2.2.3 Generic, yet semantically flexible: referential construals of implicit objects

44This subsection discusses generic implicit objects, showing how generic meaning is lexically and contextually determined. We also show, drawing on Cummins & Roberge (2004), how implicit generics, when 'strategically' placed in some discourse contexts, can receive referential construals via conversational implicature.

45An implicit object contributed by an IOC can receive a generic reading, which would be marked by a determiner if it were to be overtly expressed. The examples from COCA in (8), some repeated from (1), illustrate implicit generics, with possible interpretations suggested in parentheses:

46The interpretation of implicit objects in (8) is determined not only by the selection restrictions of individual verbs, but also by their aspectual characteristics. We discuss each set of restrictions in turn.

47Cummins & Roberge (2004), Mittwoch (2005), Pérez-Leroux et al. (2013) argue that selection restrictions determine lexical noun classes compatible with a verb's theme/patient role. In each verb’s thematic grid, a theme argument is associated with a relevant noun class, or classes, if there is more than one. Based on the meanings of the implicit objects in (8), as well as the previous examples, it can be further proposed that the relevant class is designated in the thematic grid by a superordinate term (hypernym), which is connected to a range of hyponyms. In other words, a selected noun class is organized as a hierarchical hyper-hyponymic semantic network, with some hyponyms used in lieu of a superordinate term.

48For example, eat and play can be described as having two prototypical semantic classes associated with their theme arguments, shown in the examples in (9), which illustrate hyper-hyponymic relations:

50These examples show that even though ‘food’ is the most obvious meaning of the implicit objects, the information it conveys varies by subject argument, and is enriched by an activated network of hyponyms through the general cognitive process of inferencing. (Note that if someone does not know what rabbits eat, let alone camels, the inferencing process would generate a semantic representation limited to a superordinate term.)

51Aspectual characteristics of predicates instantiating an IOC contribute a separate layer of meaning to implicit object semantics. At the level of lexical aspect, verbs instantiating an IOC are activities and represent an atelic aspectual type. At the level of grammatical aspect, they are marked as habitual/generic, encoding unlimited repetitions of a given eventuality over an unbounded time span. As Mittwoch (2005: 247) points out, habitual aspectual semantics, if realized by a verb allowing an implicit object, also provides for “an unlimited number of instantiations of the denotee of the missing object.” Therefore, a generic interpretation of implicit objects is derived from the specifics of habitual aspect semantics, which makes them non-referential, with a denotation of semantic type.

52A dialogue in (11), our own, shows that an implicit object in habitual sentences introduces no discourse referent that can be ‘picked up’ by a pronoun:

53The object argument of read in (11) can be understood as a semantic network, which includes ‘published material: books, newspapers, research papers, etc.’; yet, replacing any of the hyponyms (or a hypernym) with a pronoun is an impossibility because the theme argument has not been assigned neither a syntactic relation nor a discourse function. While it is obvious that an anaphoric chain based on implicit generics cannot be constructed in (11), it is nevertheless possible for implicit objects to form anaphoric chains with preceding antecedents and inherit their semantics (i.e. be antecedent-dependent). As illustrated in (8c), the implicit object of ‘build’ is unambiguously homes, based on the preceding sentence that introduces the topic of home-building into the discourse ('The people building your home will have opinions –after all, they build for a living'). Note that homes can also be analyzed as a vernacular hypernym, subbing for the cognate noun buildings, which is also an appropriate option for this implicit argument.

54The property of implicit generics, allowing linking to a referential antecedent, is discussed in Cummins & Roberge (2004), who give the following example, shown in (12):

55In (12), Speaker B answers Speaker A's question and therefore her/his utterance must be interpreted as communicating relevant information. We observe two characteristics that distinguish (12) from the other examples in this section: (i) the verbs chop and dice are used in present perfect progressive and non-habitual/episodic aspect; (ii) it is possible to construct two meanings for the implicit object of chop and dice: a generic meaning with a single or multiple denotees (vegetables (carrots, onions, mushrooms, etc.) and/or fruits: apples, oranges, etc. and/or herbs: basil, oregano, sage, etc.) and a referential/definite meaning (the vegetables). With the referential construal, the implicit object semantics is antecedent-dependent and requires some inferencing on Speaker B's part along the following lines: “if Jacques has been busy in the kitchen chopping and dicing (vegetables and other things), he must have chopped and diced all of the vegetables we have (otherwise, I would be able to find them).” While Cummins and Roberge do not provide an extensive discussion of this interesting example, they do point out that the referential interpretation is not "forced," implying that it is only available as an alternative. In sum, (12) illustrates that propositions with generic implicit objects can receive an antecedent-dependent interpretation, while generating an implicature that includes a definite/specific referent as an alternative denotation. In Section 3, we introduce our study, which investigates context-based referential and non-referenetial construals of an IOC.

3.1. Goals

56As we observed earlier, experimental studies of implicit objects have been mainly conducted by first and second language acquisition researchers. This study aims to explore speakers' acceptability judgments of an implicit object construction in two types of discourse contexts:

57(i) in the context of 'low discourse prominence,' when an implicit (de-profiled) object receives generic/non-referential construal, and its meaning is derived in the absence of an antecedent, lexical or deictic/pragmatic (this context type 'licenses' the De-profiled Object Construction);

58(ii) in the context where an overt [+referential] antecedent is available and can therefore contribute to the interpretation of an implicit object, with the latter receiving referential construal via the mechanism of conversational inference (Cummins & Roberge 2004).

59In both context types, speakers' acceptability judgments of implicit objects are evaluated in relation to the versions of a counterpart construction with a lexicalized object (see Section 3.3 for details of the experimental design).

3.2 Hypotheses

60The following three hypotheses were formulated:

61(i) in contexts of low discourse prominence for an object argument, an IOC is predicted to receive higher acceptability rankings than its counterpart with a lexicalized object (expressing a generic object should be viewed as informationally superfluous, resulting in degraded acceptability);

62(ii) in [+antecedent] contexts, an IOC is predicted to have lower acceptability rankings than its counterpart with an expressed object (a construction with an overt object, expressed by a pronoun clearly linked to an antecedent, is not ambiguous and therefore is predicted to be ranked higher);

63(iii) in contexts of low discourse prominence, an IOC is predicted to receive higher acceptability rankings than in contexts where it can be linked to an overt antecedent (in the latter, an implicit object is semantically ambiguous, with a referential meaning triggered by conversational implicature).

64In the following section, we introduce the experiments.

3.3 Experiment 1: Acceptability Judgments

3.3.1 Participants, stimuli, procedures

65Thirty one monolingual native speakers of American English (ten females, twenty one males) participated in the first experiment. The mean age of participants was 29.48 (age range: 19-54; SD 7.14). In a brief background questionnaire, the participants were asked to indicate their age at the time of the study, their native language (all listed American English), other languages they grew up speaking (all answered ‘none’ to this question), and their level of education at the time of the study (high-school (1), BA/BS (23), MA/MS (6), PhD (1)). The participants were recruited through the Reddit social platform. They were not compensated for taking part in the study.

66All 31 speakers completed an online task that included 36 dialogue items (18 experimental and 18 distractors). A dialogue item consisted of a brief exchange between Speaker A and Speaker B. Speaker’s B response was presented in a column format that listed three alternate sentences, and the participants ranked each alternate on a seven-point scale. (Each participants completed the entire experiment; there were no missing observation points in the data.) The instructions were as follows:

Your task is to determine which response (out of 3) is the best fit for an opening line. Rank each of the given responses, using a seven-point scale (1/2/3/4/5/6/7). One (1) means ‘very bad’ and seven (7) means ‘very good.’ The numbers in-between indicate various levels of ‘goodness’ and ‘badness.’ In some cases, you may prefer only one response. In other cases, more than one might feel like a very good fit.

It is VERY IMPORTANT to read each response and an opening line TOGETHER, as an ENTIRE dialogue, from the beginning to the end, BEFORE choosing your ranking.

ALWAYS RELY ON YOUR FIRST IMPULSE. Sometimes you may feel that more than one response is appropriate and could be used in informal speech, and this feeling is NORMAL. These exchanges are meant to be read as if they were spoken by people (NOT as examples of written, formal English).

67An example of an experimental dialogue is shown in (13). (The ranking scale, which consisted of 7 star-shaped points, appeared on the same line as an alternate response.)

69The Generic Condition consisted of the following implicit/overt objects alternations:

71The Indefinite Condition tested whether an IOC can trigger an implicature to accommodate a specific indefinite referent into its denotation, thus allowing referential construal as an alternative to generic interpretation. If so, an IOC should also be felicitous in a [+antecedent] context, which should be reflected in its acceptability rankings. Speaker’s A opening line included an antecedent that an implicit object could be linked to. The three alternates for Speaker’s B response were (i) an IOC, (ii) its replica with an overt pronoun ‘it,’ which anaphorically linked to an antecedent in the opening line, and (iii) an alternate with an overt noun (mainly the indefinite pronoun ‘something,’ which was somewhat infelicitous in [+specific] contexts). A third least felicitous alternate was included in the set to reduce the odds of guessing and to ensure that participants considered all three responses in their rankings, but it was not a variable of interest in the study. All of the antecedents in Speaker A’s line were singular count nouns and each antecedent was compatible with the pronoun ‘it’ in an anaphoric chain. Example (14) shows an experimental item from the Indefinite Condition:

72The Definite Condition included a definite/specific noun in Speaker A’s opening line. This condition tested whether an IOC would be acceptable if an implicature accommodating a definite referent into the implicit object's denotation were to be triggered. The alternates for Speaker’s B response included (i) an IOC, (ii) a sentence with the pronoun ‘it’ as an object argument, and (iii) an alternate with the indefinite pronoun ‘something,' which was infelicitous in the [+definite antecedent] context. All of the antecedents in Speaker A’s line were singular nouns. Example (15) shows an experimental item from the Definite Condition:

3.3.2 Results

74Descriptive statistics for the mean rankings of the implicit object constructions in all three conditions are reported in Table 1. Figure 3 summarizes the distribution of the participants’ rankings.

75Table 1. Mean rankings of the de-profiled (implicit) object construction

76The results indicate that the participants give the highest rankings to the IOC in the Generic condition (mean 6.77), with a SD deviation of 0.55 suggesting uniformity in the acceptability pattern. (The rankings of the implicit objects do not fall below 5.) The IOC shows a different pattern in the [+referential antecedent] contexts. While the mean rankings of 5.15 (for the Indefinite Condition) and 5.02 (for the Definite Condition) indicate overall acceptability of the construction, it has much greater dispersion on the acceptability scale. There are participants who clearly reject this construction in some contexts, as shown by the rankings in the 1 to 2 range. (On the seven-point scale, the 1-to-3 range indicates low levels of acceptability, with the marks of 1 and 2 being interpreted as outright rejections.)

77Paired-samples t-tests, with the Bonferroni adjustment method, were conducted to compare the mean rankings of the implicit object responses. Table 2 summarizes the findings.

78Table 2. Pairwise comparisons of the implicit object responses

79As shown in Table 2, there was a significant difference (p < .001) in acceptability between the Generic Condition and the Indefinite Condition (6.77 vs. 5.15), as well as between the Generic Condition and the Definite Condition (6.77 vs. 5.02). This difference suggests that the IOC has much greater acceptability in the contexts in which an implicit object is construed as a ‘pure’ generic and in which it is not "forced" to include an implicature-based denotation for a specific referent as an interpretive possibility. The difference in the mean rankings between the Indefinite and Definite Conditions (5.15 vs. 5.02) was not significant, suggesting that the (in)definiteness-based differences between the antecedents (specific indefinite vs. specific definite) is not a factor in its acceptability ratings.

80Descriptive statistics for the mean rankings of the overt object response (lexical generic nominals and referential ‘it’) are reported in Table 3. Figure 4 summarizes the distribution of the participants’ rankings.

81Table 3. Mean rankings of the overt object responses

82The mean rankings suggest that the overt-object response has overall high mean acceptability in all three conditions. In the Indefinite Condition, it shows the highest value (6.65), with the highest concentration of rankings hovering around 7. In the Definite Condition, most of the responses cluster in the 6-to-7 range. The greatest dispersion of the responses is in the Generic condition, with 50% clustering in the 5-to-7 range.

83Paired-samples t-tests, with the Bonferroni adjustment method, were conducted to compare the mean rankings of the overt object responses. Table 4 summarizes their results.

84Table 4. Pairwise comparisons of the overt object response

85As shown in Table 4, there was a significant difference (p < .0005) in acceptability between the Generic Condition and the Indefinite Condition (5.89 vs. 6.65), as well as between the Generic Condition and the Definite Condition (5.89 vs. 6.24). These differences suggest that the availability of an antecedent is a factor in the selection of the overt object alternate and its higher rankings. A comparison of the Indefinite and Definite Conditions also revealed a significant difference (6.65 vs. 6.24), suggesting that the overt alternate with referential ‘it’ is ranked higher in the Indefinite than the Definite Condition. This finding indicates that the (in)definiteness status of an antecedent is a factor in the acceptability rankings of the object arguments expressed by referential ‘it.'

86Paired t-tests were also conducted to compare the mean rankings of the implicit object responses with their overt alternates within each condition. Table 5 summarizes the findings.

87Table 5. Implicit objects vs. overt objects (lexical generic nouns and referential ‘it’)

89Turning to individual verb patterns in each experimental condition, our analysis revealed the following gradation of predicates in sentences with implicit objects (see Appendix I for tables with individual verb data):

Generic Condition (means are in parentheses):

explore (6.83), eat/park/play (6.77), cook (6.74), clean (6.7)

Indefinite Condition (means are in parentheses):

play (6.35), park (6.06), clean (6.03), eat (4.87), explore (4.35), cook (3.22)

Definite Condition (means are in parenthesis):

play (6.77), park (5.93), clean (5.22), explore (5.19), eat (4.09), cook (2.90)

90In the Generic Condition, the means of individual verbs are quite uniform, suggesting that all the predicates used in the study are acceptable when their implicit objects are interpreted as unambiguously generic. By contrast, there is much greater variability in the acceptance means in the [+antecedent] conditions, with the verb 'cook' receiving the lowest rankings and 'play/park/clean' the highest. These data suggest that referential construals of implicit objects (derived by an implicature) may vary by verb's lexical semantics and/or the extent to which it 'participates' in the IOC. We address this issue in Section 4.

3.4 Experiment 2: A Forced-Choice Task

91The goal of Experiment 2 was to see if the tendencies in the acceptability judgement task will be corroborated by a different task. How accurate were the speakers' rankings in Experiment 1? Will the same response patterns emerge if the stimuli were presented in a multiple-choice format and participants were asked to choose only one sentence out of three?

3.4.1 Participants, stimuli, procedure

92A different cohort of twenty-four monolingual native speakers of AE (twelve females, twelve males) participated in the second experiment. The mean age of participants was 27.5 (age range: 18-66, SD 12.6). In a brief background questionnaire, they were asked to indicate their age at the time of the study, their native language, what other languages they grew up speaking (all answered ‘none’ to this question), and their level of education at the time of the study (BA/BS (19), MA/MS (5)). The participants were recruited through the Reddit social platform. They were not compensated for taking part in the study.

93The participants were given the same set of stimuli as in Experiment 1, but the sentence evaluation component of the task was restructured as a forced multiple-choice. The instructions asked to choose only one response (out of three), viewed as the “best fit” for Speaker A's opening line. Each participants completed the entire task; there were no missing observation points in the data.

94Qualtrics software was used to create an online version of the experiment. All the items, experimental and distractors, were randomized for each participant. The alternates for Speaker B’s responses were also randomized, both for the experimental and distractor dialogues.

3.4.2 Results

95The results are reported in Figure 5:

96The forced-choice version of the task corroborated the rankings reported in Experiment 1. In the Generic Condition, the IOC was the preferred response (89.3%, SD 0.106), followed by the overt generic object alternate (10.7%, SD 0.11), with the third alternate (which included mainly the indefinite pronoun ‘something’) receiving no selections. In the Definite and Indefinite Conditions, the alternate with referential ‘it’ was clearly the preferred response (75%, SD 0.23 vs. 72.6%, SD 0.20 respectively). The IOC was selected 25% (SD 0.23) vs. 22.6% (SD .20), suggesting that some speakers preferred the implicit object construction over the overt object. The alternate with the indefinite ‘something’ was hardly ever selected (0% in the Definite Condition vs. 4.8%; SD 0.078 in the Indefinite Condition). Thus, the results of Experiment 2 corroborate the tendencies for the preferred responses reported in Experiment 1.

97In this study, we investigated American English speakers' acceptability of the IOC in two types of contexts, comparing the low discourse prominence context, which favors generic interpretations, with the context in which an implicit object meaning can be antecedent-dependent. In the latter, referential construal is derived via conversational implicature as an IOC can be (optionally) interpreted in relation to an overt [+referential] antecedent. In the CxG, a construction with a de-emphasized object is analyzed as a De-profiled (Implicit) Object Construction (Goldberg 2001, 2005), with an internal structure that assigns no pragmatic function (topic/focus) and no syntactic relation to a theme participant of the predicate. A lexical verb that instantiates a DOC is argued to ‘inherit’ a theme participant semantics from the construction, which makes for its obligatorily transitive interpretation, even though it is solely a verbal predicate that is 'emphasized' by the construction type. In addition, we presented empirical evidence from the literature and from COCA, showing that implicit objects can be referentially construed (as specific indefinite or definite) in contexts where lexicalized arguments of the predicate and/or conversational topic make their semantics 'highly predictable' (cf. Fillmore 1986; Mittwoch 2005; Rakhilina & Plungian 2010). In Cummins & Roberge (2004), we also found uses of an IOC in a referential sense, with an implicit object being anchored to an antecedent via conversational implicature. To the best of our knowledge, AE speakers' acceptability of an IOC with implicature-based referential vs. generic construals, as well as its acceptability in comparison to overt objects (referential vs. generic) has not been previously studied.

98In the acceptability judgments experiment, an IOC was investigated in three experimental conditions: Generic (no antecedent), Indefinite (+ specific indefinite antecedent), and Definite (+ definite antecedent). Six verbs were randomly selected as predicates for an IOC and were used in all three conditions: clean, cook, eat, explore, park, play. There was a statistically significant difference (p < .001) in the acceptance rankings between the implicit objects with a generic interpretation (Generic Condition) and the implicit objects with an implicature-based reference (Indefinite and Definite Conditions). As shown in Figure 3, the implicit generics rankings were restricted to the 5-to-7 range, with the mean of 6.77 (there were no rankings in the 1 to 4 range). By contrast, there was much more variability in the speakers' judgements of implicit objects in the [+antecedent] conditions. The acceptance rankings were dispersed across the entire 7-point scale. In the Indefinite Condition, 50% of the data concentrated in the 4 to 7 range (with the overall mean of 5.15). In the Definite Condition, 50% of the data clustered in the 3 to 7 range (with the overall mean of 5.02). This finding provides support for the hypothesis that 'de-profiled' implicit generics should receive higher rankings vis-à-vis uses of an IOC that (optionally) allow referential, antecedent-based construals.

99From the perspective of the CxG, this difference in speaker acceptability can be accounted for terms of conventionalization of function. A de-profiled construction with implicit generics is highly conventionalized in its usage, with lexical verbs contributing 'default' (lexically presupposed) meaning to their object argument, and speakers accessing these meanings in the comprehension process. The use of an IOC with a referential implicature constitutes a departure from this conventionality as the construction meaning gets flipped from the generic to referential realm, provided an antecedent-based interpretation is accessed. In our data, there was no statistically significant difference between the Indefinite and Definite Conditions in speakers' acceptability of an IOC, suggesting that the (in)definiteness status of an antecedent was an unlikely influence in their judgments. However, the analysis of individual verb data can provide some insights into why speakers' acceptability of an IOC dropped in the [+antecedent] contexts. Our data revealed some variation in the extent to which the speakers accepted each of the six verbs used in the study. There was an interesting gradation in the mean acceptance rankings of individual verbs, with the rankings spread across the 7-point scale: cook received the lowest rankings of 3.22 (Indefinite Condition) and 2.90 (Definite Condition), which suggest speakers' uncertainty, if not outright rejection. The highest rankings were given to play, park and clean (5.22 to 6.77 range), with explore and eat falling in the mid-range (4.09 to 5.19). In the CxG, this variability in speakers' judgments receives a natural explanation if some verbs 'gravitate' towards an IOC more than others and if some verbs instantiating an IOC gravitate towards [+antecedent] contexts more than others. Clearly, the issue of how the semantics of implicit objects vary by individual verb and [+/-antecedent] context needs to be explored in future experimental and/or corpus-based studies, especially in light of our finding that the acceptability rankings were strikingly uniform for all of the verbs in the Generic Condition.

100Turning to the analyses that compared an IOC to its variant with an overt object, the findings show statistically significant differences between the two constructions in each experimental condition. The AE participants consistently ranked the De-profiled Object Construction higher than its alternate with a lexicalized generic NP (cook/meals, clean/things, play/cards, park/a car, explore/places, eat/food). The reverse pattern characterized the [+antecedent] conditions: the AE participants consistently ranked the construction with the referential pronoun 'it' higher than the IOC. As shown in Figure 4, in the Indefinite Condition, most of the overt-object responses clustered around the highest point of the scale. In the Definite Condition, 50% of the responses clustered in the 6-to-7 range. By contrast, in the Generic Condition 50% of the overt-object responses concentrated in the 5-to-7 range. One statistically significant difference emerged between the Indefinite and Definite Conditions: referential 'it' was more readily accepted in the Indefinite than the Definite Condition (6.65 vs. 6.24, p < .001). Thus, in our study, the in(definiteness) status of an antecedent influenced the rankings of pronominal objects (when these appeared as an alternative to an IOC). While more work is needed to understand how the semantics of antecedents contributes to the choice of overt vs. implicit objects, we tentatively suggest that the lower acceptance of 'it' in the Definite Condition could be a function of speakers' expectations to 'see' a greater variety of truncated structures. In other words, alternates with an overt pronoun might not be the best alternative structure for [+definite] contexts. For example, in AE spoken dialects, it is common to drop nouns (not just objects) in 'highly predictable' phrases: e.g. 'And the last four _ of your social _?' (with the nouns 'digits' and 'number' omitted). In future work, it might be interesting to explore other, more colloquial alternatives to an IOC in contexts with [+definite] antecedents.

101Experiment 2, which used the same stimuli but redesigned Speaker B's responses as a forced multiple-choice task, corroborated the main findings of the acceptability judgements study. A different cohort of participants completed the multiple-choice version, and their response patterns aligned with the preferred response patterns in the ranking task. The participants consistently chose the implicit objects in the Generic Condition and the overt pronominals in the [+antecedent] conditions. A combination of the acceptability and multiple-choice tasks, using the same stimuli but different participants, has been shown to be a useful and informative methodological strategy.

102The findings reported here contribute new experimental evidence to the study of the implicit object phenomena. The CxG has been shown to provide a fruitful approach to understanding the nuances of semantic/pragmatic properties of IOC(s) and their diverse patterns of usage (for some general discussion of the semantics/pragmatics interface in the CxG see Leclercq 2020, to appear). This interesting topic can be explored in future research using different experimental tasks, different verb classes, and different discourse conditions. Last but not least, it remains to be seen how the IOC patterns AEin other dialects of English (Australian, British, Canadian, etc.), as well as in other languages. As emphasized in Östman & Trousdale (2013), the CxG approach can be effectively used to shed light on intra- and inter-dialectal variation.

103This study explored discourse-based variability in speakers' acceptability of an implicit object construction in American English. The analyses from the acceptability judgment experimentand the multiple-choice experiment, which used the same stimuli but were administered to two different cohorts of participants, converged on two main findings. An IOC with generic interpretation was judged to be most acceptable in discourse contexts with no antecedent for an implicit object, that is, contexts, which triggered interpretations of object arguments mainly determined by individual lexical verbs and their thematic grids. The speakers' acceptability of an IOC was degraded in contexts where implicit objects could be construed referentially (that is, contexts that provided an overt lexical antecedent for an implicit object) and where an alternate structure with an overt object was available as a possible choice. Nevertheless, there was considerable variability among participants in the acceptability of an IOC in [+antecedent] contexts, as well as variability by individual verb that instantiated an optionally referential use of the construction. Future studies can further test whether our findings are generalizable to other dialects of English and whether similar patterns can be found using alternative experimental methods.

104Appendix I: Individual Verb Analysis by Experimental Condition

105Table A. Individual verbs’ mean responses in the Generic Condition

106Table B. Individual verbs’ mean responses in the Indefinite Condition

107Table C. Individual verbs’ mean responses in the Definite Condition

108Appendix II: Linear Model Analyses Based on Residualized Ranking Data (Diviak et al. (2016)

109Table D. Within-condition comparisons: results of a linear regression model

110Table E. Between-condition comparisons: results of a linear regression model